techniques in plasma diagnostics richard a. gottscho terry...

Pure Appi. C7zem., Vol. 56, No. 2, pp. 189—208, 1984. 0033—4545/84 $3.OO+O.OOPrinted in Great Britain. Pergamon Press Ltd.

©1984 TUPAC

Optical Techniques in Plasma Diagnostics

Richard A. GottschoTerry A. Miller

Bell LaboratoriesMurray Hill, New Jersey 07974

ABSTRACT

Recent developments in optical techniques for plasma diagnostics are critically reviewed. Theprimary emphasis is on plasma-induced emission and laser-induced fluorescence probes of radical andionic concentrations and temperatures in low pressure discharges like those used in microelectronicfabrication. Other techniques such as optogalvanic, infrared, spontaneous and stimulated Raman, andmultiphoton spectroscopy are also discussed briefly.

Generally, emission techniques are seen to be useful primarily for qualitative analysis andcharacterization of the electron energy distribution. However, in some systems, emission actinometry,where an inert gas is used to account for changes in electron density, appears to be a valid means ofdetermining relative changes in concentration. Laser-induced fluorescence spectroscopy has only recentlybeen utilized for in situ measurements of ground state ions and radicals but has already yielded insightinto mechanisms for formation, destruction, energy transfer, and maintenance of the discharge. By andlarge the other techniques show great promise but remain to be exploited as diagnostic tools for plasmas.

I. INTRODUCTION

Diagnostic techniques for plasmas have shared in the growth in interest and popularity that thegeneral field of plasma chemistry and processing has recently enjoyed. This is true even though the goalof plasma diagnostics is the basic understanding of plasma chemistry and related phenomena, ratherthan the production of an economically viable device or material. The key to this overall interest is thatthe success of a particular process leading to a desired product can often depend critically on the delicatebalancing of numerous factors inherent in the plasma. For example in the production of micro-electronicdevices, factors that must be considered include, surface bombardment by positive ions - their energyand directionality, tvailability of neutral reactive species, production and destruction of surface films onwafers, etc.

For most plasma processes a number of variables, e.g. gas mixture, gas flow, rf power, etc., can beexternally controlled to optimize the balance between competing chemical and physical processes.However the number of variables available is so large that the employment of a random samplingtechnique to achieve optimization of reactor conditions for a given process would place severe demandsupon the patience of any investigator. For this reason it is important to develop reliable diagnostic tools,both to provide a basic understanding of the plasma and to serve as useful indicators for actualprocesses.

It may be convenient to divide diagnostic techniques into three general categories, ex situ, in situ-intrustive, in situ-non-intrustive. The common characteristic of ex situ techniques is that they sample analiquot of the plasma reactor's contents and transfer it elsewhere for examination. Analytical techniquesthat have been employed in this manner are mass-spectrometry, gas-phase electron paramagneticresonance, gas chromatography, and chemical titrations via downstream reactions. The advantage of theex situ techniques is that they allow the full power of numerous analytical techniques to be employed.The chief disadvantage of ex situ techniques is fairly obviously. Since the contents of a plasma reactorencompass many highly reactive species, ions, free radicals, reactive atoms, one can never be certain thatspecies analyzed ex situ have not been transmuted during their journey from the reactor.

The division of in situ techniques into intrustive and non-intrustive is somewhat arbitrary. To someextent, any diagnostic technique perturbs the plasma. However, in some cases this perturbation may beso slight as to be negligible, while in others it may be quite considerable. One example of an in situ

189

190 R. A. GOTTSCHO and T. A. MILLER

diagnostic which would generally be called intrustive is a Langmuir probe. Clearly in this case one isinserting a physical object into the medium and thereby perturbing it. However, depending upon theinformation desired this perturbation may or may not be negligible.

Almost certainly the least intrustive in situ diagnostics yet developed use photons as the informationtransmitter between the medium and the observer. Rarely will they perturb the medium in a non-negligible manner. Such photon-driven diagnostics, conveniently called optical diagnostic even thoughthe photons involved may be outside the spectral region visible to the human eye, will be the subject ofthe remainder of this paper. We will review early optical diagnostic work which consisted mainly of ananalysis of the optical emission generated by the plasma itself. We will discuss attempts to make thistechnique more quantitative by the use of actinometers. The success and limitations of actinometrictechniques are considered.

We also discuss laser induced fluorescence diagnostics of plasma systems. We consider some of theadvantages of laser induced fluorescence especially in the areas of high spatial and temporal resolution,and quantitative analysis, and we illustrate the kinds of plasma information obtainable from suchtechniques by the use of specific examples.

We will also touch upon some of the other "optical" techniques that have possible application forplasma diagnostics. These include opto-galvanic techniques, IR emission and absorption including lasertechniques, Raman scattering, and multi-photon excitation. We conclude with a short summary and alook towards future applications. This paper is written from the optical diagnostic or spectroscopic pointof view. In this sense, the techniques described could be applied to either a thermal plasma or a non-equilibrium plasma. Our own work is in the non-equilibrium plasma area and it is for this reason thatthe examples we cite are from this area.

II. EMISSION SPECTROSCOPY DIAGNOSTICS

Most plasma reactors give off a variety of "optical" emissions ranging throughout the spectrum fromthe IR to the UV. One of the simplest diagnostics for plasmas is the spectral analysis of this radiation.Usually this is accomplished by passing the emission through a monochromator and recording its timeaveraged spectral features. It is convenient to illustrate the advantages and limitations of emissiondiagnostics by recounting the history of its application to one of the most widely studied plasma systems,the etching of Si and SiO2 by a CF4-02 plasma.

A. CF4-O2 Emission

A little over 5 years ago, Harshbarger, Porter, Miller, and Norton (1) undertook a simple study ofthe resolved emission from a standard parallel plate plasma reactor etching Si with a mixture of CF4-02.Fig. 1 shows a typical emission spectrum recorded with Si wafers present (bottom) and absent (top). As

Wavetengh (nM)

Fig. 1. Emission spectrum of CF4-02 plasma. Bottomtrace Si wafers present, top trace absent.

Optical techniques in plasma diagnostics 191

Fig. 1 shows, the three principal species emitting from the CF4-02 plasma are F and 0 atoms and theCO molecule. The trace with Si wafers shows drastically less atomic emission but comparable COemission. From this evidence alone, one would likely conclude that the F and 0 atoms are active in theetching process but CO is not.

In Fig. 2 we have plotted the F atom emission, the 0 atom emission and the Si etch rate vs. 02

concentration. We note that the F atom emission and etch rate are strongly correlated but do not havean identical functional form. The 0 atom emission shows little direct relationship with the Si etch rate,but increases rapidly as a function of 02 once the maximum in the Si etch rate is reached. A modelqualitatively consistent with the above data can be drawn as follows. F atoms are the active speciesetching Si. When small amounts of 02 are added to the discharge, chemical reactions involving 02and/or 0 atoms liberate more F atoms, which results in an increase in the etch rate. The reaction of 0atoms with some species in the reactor controls the concentration of 0 atom emission, till a point nearthe maximum Si etch rate when this reaction is no longer efficient enough to prevent a rapid increase inO atom emission.

C

E

EC

w

UI—LU

Cl)

These qualitative conclusions based upon the simple emission diagnostics are all still believed to bequalitatively correct, but they are not the whole story. For instance, Mogab, Adams and Flamm (2,3)carried out a more detailed investigation of the CF4-02 etching of Si and confirmed that while thecurves giving F atom emission and Si etching vs. 02 concentration are similar, they are certainly notidentical. This result is not consistent with the assumptions that (i) the emission intensity, 'F' isproportional to the F atom concentration [F] and (ii) that the etch rate R is proportional to [F]. Forthen

where

R = ($F/c$) 'F

R=3F[F]with 13F = proportionality constant relating R and [F] and

[F] = IF/aF

with a= proportionality constant relating 'F and [F].

(1)

(2a)

(2b)

Let's first consider the assumption that 'F and [F] are proportional, by examining the proportionality

constant a in more detail,

o to 20 30 40 50% 02

Fig. 2. F, 0 emission, etch rate vs. 02 addition (seeRef. 1).


c$ = KL'

(P,Ne) Ne() th (3)

where

K = a constant depending on the sensitivity of the detector.

= cross-section for excitation of the emitting species (F atoms)to a given excited state by impact of an electron of energy .

Ne(f) = number of electrons in the energy range dfpresent in the volume of the reactor viewed by the detector.

Q (P,Ne) quantum yield for emission from the given excitedstate of interest as a function of discharge pressure andelectron density.In principle Q varies between 0 and 1.

When one looks at the factor a in this manner, it is obvious that it is anything but a constant. Infact it is remarkable that an equation like (2b) has any usefulness at all. Rigorously speaking, thenumber density of electrons as a function of energy and the quantum yield must be invariant for theemission and concentration to be simply related. Since Ne() depends on applied RF power and gasmixture and Q depends on total pressure and gas mixture, $ will never be precisely a constant.Furthermore Eq. 3 assumes that all the F emission originates from excitation of ground state F atoms.If other processes contribute to 'F' e.g. CF4 + e — CF3 + F , then Eq. 2b is totally invalid.

However in some cases of practical interest things may be somewhat better than indicated above.For plasma etching applications, the pressure and electron density are relatively low, thus Q(P,Ne) maybe very nearly unity over a range of operating conditions, especially for short-lived excited states.Similarly cr) may be a relatively "slow' function of , and Ne(() may not depend extremely strongly ongas composition. Nonetheless, variations in a due to these considerations are nearly impossible to avoid.Similar, but more difficult-to-estimate, are the errors likely introduced by processes contributing to 'F'other than ground state F atom concentration excitation.

In the work of Mogab, et al., (2) an effort was made to measure the variation of a as a function of02 addition to the discharge. For this experiment an independent method for [F] determination isrequired and these workers employed a downstream chemical titration of F atoms. Such an ex situtechnique eliminates the possibility of an absolute calculation for a, but barring some chemicalpathologies, it is plausible to assume that the relative variation of a with varying 02 concentrations canbe measured. The results of these experiments are shown in Fig. 3.

The value of a can be seen to vary by over a factor of two for a reasonable range of 02 addition.While this change is quite modest compared to what could be expected from an analysis of Eq. 3 it iscomparable to the deviation between the 'F curve and etch rate curve vs. 02 concentration.Unfortunately this variation in a is precisely in the wrong direction to make the 'F and R curvescoincide!

Thus there must be something more complicated in the etching mechanism, requiring more than thesimple relation Eq. 2a between etching and F atom concentration. It has been proposed (2) that besidesits role in liberating more F atoms to etch, 02 addition produces 0 atoms which compete with F atomsfor surface sites. When 0 atom concentrations become significant, such competition can decrease theoverall etching rate. Quantitatively then

R = fl[F]2/([F1 + k[0]) (4)

where k gives the competitive efficiency for 0 vs. F atoms. It is found that by using 'F' lo and theexperimentally determined variation of $, the etch rate R can be correctly predicted by Eq. (4) over awide range of 02 concentrations (2).

The moral of the story is that plasma mechanisms can be complex and reliable diagnostics areabsolutely necessary if one is to even hope for an understanding of them. Simple emission spectroscopyis a good qualitative tool, but great care must be exercised in its quantitative application.

B. Actinometry

The non-constancy of a in Eqs. () and (2b) is quite troubling. In the above CF4-02 example, thevariation in a was determined by an ex situ chemical titration, hardly a universal solution. A more


general approach to this problem has been the development of actinometry. In this case one monitorsnot only the emission I from the reactive species of interest, but also the fluorescence, I of an inert gas,e.g. Ar or N2, in the same plasma. Then we have

= (N/N)(a/) = (Nr/Ni) X

SQr (P,Ne) o(() Ne()(5a)

SQ (P,Ne) ofk) Ne ()

It is clear that if the bracketed integrals are equivalent, then one has a very simple result. Therequirement for this equivalency is that the quantum yield and the cross sections have the same energydependence. Since for short-lived excited states in a low pressure cold plasma Q is essentially unity, thisrequirement simply reduces to one on the cross section. In no case will two different species have exactlythe same functional form for the energy dependence of the cross section, but if the actinometer species iswisely chosen, the form of and may be similar enough to give practical results of considerable

utility.

Coburn and Chen (4) tested the actinometric technique by monitoring I/Ir in CF4/02/Ar plasmasas a function of the 02 feedstock composition. They found excellent agreement between Ir and a,determined previously by Mogab et a!. (2) (Fig. 3). Thus, Eqn. (5a) simplifies to

I/Ir = KNF/NAr (5b)

where K is a constant. Ar satisfies at least two key actinometric criteria: it is relatively inert and theexcited state involved lies at —13.5 eV compared to —14.5 eV for the corresponding F state. Althoughthe technique appears to work well in this case, the origin of the F atom emission was not determined;so, the question lingers as to whether or not the agreement observed by Coburn and Chen(4) wasmerely fortuitous.

1 1

CF4 + 02

o 0 Ar (7504) INTENSITY -

EXCITATION EFFICIENCY

00

I I0 20 40 60

OXYGEN PERCENTAGE

Fig. 3. Variation of a and Ar* emission intensity (at

750.3 nm) with 02 concentration (Refs. 2 and 4).

The actinometric technique was developed independently and applied further by d'Agostinoet a!. (5-8). In studying CF4/02 plasma etching of Si and 5i02, Ar and N2 emission were used to monitorchanges in electron density as a function of 02 concentration. Despite the difference in threshold, N2second positive emission (—11.3 eV) and Ar emission (—13.5 eV) exhibited the same dependence upon02 concentration. In addition, the intensity ratios of several F lines and of CO to CO+ emission(covering an energy range from 8 to 20 eV) did not change when up to 50% 02 was added to a CF4plasma. This suggests that the electron energy distribution was also invariant and that the primary


electronic change which occurs when 02 is added to a CF4 plasma is a decrease in charge density. Thisis manifest by a decrease in excitation efficiency (Fig. 3). After so characterizing the energydistribution, d'Agostino et al., used N2 and Ar emission to normalize F, 0, CO, and CO2 intensities andobtain estimates of relative changes in concentration, assuming that the emission in each case arose fromelectron-impact excitation of the ground-state species. Ar and N2 were used again as actinometers in astudy of SF6/02 plasmas (6); but, in this case, the technique was validated ex situ by downstreamtitration measurements.

Donnelly et al. (9) also validated the use of Ar actinometry by downstream titration when theymeasured F atom concentrations in CF4/02/Ar and NF3/Ar plasmas in determining the Si02 etchingmechanism. For F atom densities from 1014 to 1016 molec cm3, Eqn. (5b) was found to hold.

In a non-intrusive in situ test of the actinometric technique, Ibbotson et al. (10) compared 'r"Xr tothe Br atom concentration in a Br2/Ar plasma. The Br atom concentration was deduced from the Br2molecular concentration, which was measured in turn by optical absorption. This technique assumesthat each dissociated Br2 corresponds to two Br atoms. Normalized emission ('r/'r) and concentration(NBr/Ntot) measurements showed the same dependence on frequency over the range from 0.1 to14 MHz. This is a noteworthy result since changes in frequency over this range can have profoundeffects on the electron density and energy distribution.

In studies of CF/H2 plasmas, actinometry was pushed further when CF and CF2 radical emissionwas normalized to Ar and N2 emission (7). In this case, however, no evidence was presented to supportthe assumption that CF and CF2 emission comes primarily from direct electron impact on the radicalground states. Tiller et a!. (ii) also applied actinometry to the measurement of CC1 radicals producedby plasma decomposition of CC14 without verifying that CCI* comes from electron-impact excitation ofCC! ground states.

(I)

cr

>.-I-U)zwF-

(1.-Xi4)

0.7

Fig. 4. CC1 emission (normalized toCC!4 discharges as aconcentration (top) and(bottom) (Ref. 12).

N2) and LIF fromfunction of Cl202 concentration

0 0.1 0.2 0.3 0.4 0.5 0.6


In a non-intrusive in situ study, CC! radical emission was normalized to N2 emission fromCC14/02/N2 and CC14/C12/N2 plasmas while the CC1 radical density was measured by laser-inducedfluorescence by Gottscho et al (12) (see Section III). As the percentage of 02 or Cl2 in the feedstockwas increased, the CC1 concentration decreased rapidly, the CC1 emission decreased much less rapidly,and, N2 emission remained constant (Fig. 4). The reason why N2 actinometry does not work in thiscase is evident from LIF measurements of the CC1 concentration: electron-impact excitation of ground-state CC1 is relatively improbable because the steady-state concentration is low (_10b0 molec cm3); thedominant mechanism for producing CC1* is electron-impact dissociation of CClX precursors. In general,actinometry is unlikely to be valid whenever the ground-state species is short-lived, in low abundance,and derived from a molecular parent. Thus actinometry is likely to fail for CF radicals in CFdischarges but not for CF2 radicals in the same discharges because CF2 is relatively inert and long-lived (13,14).

A relatively simple way to test the validity of actinometry (non-intrusively in situ) is to examineemission lineshapes. Because electronic to translational energy transfer is inefficient, line-widthsgenerated by electron-impact excitation are generally narrower than line-widths resulting from electron-impact dissociation. Since dissociation ordinarily occurs by excitation to a repulsive electronic state,fragment radicals often leave with excess translational energy (15). Using a Fabry-Perot interferometer,Gottscho and Donnelly (16) found that the F* translational energy distribution in a CF4/02/Ar plasmawas generally cold like that for Ar*. Over a range from 0 to 50% 02 in CF4 (with 5% Ar), the F* andAr* temperatures remained constant and equal to 360 ± 70K. These observations suggest that F* isformed primarily by electron-impact on F ground-state and should quell the lingering doubts about thevalidity of actinometry in the CF4/02/Ar system.

In Cl2/Ar plasmas, on the other hand, the Cl* and Ar* translational temperatures were equal andcold only in the plasma center. In the sheath, the Cl* emission lineshape was distinctly non-Gaussian,with a broad pedestal superimposed on a sharp, central peak. The relative intensity of the broadpedestal increased on the negative part of the cycle in these low frequency (20-100 kHz) plasmas,suggesting that an ion-molecule reaction produces some Cl*. Ar* also showed a pedestal during thenegative part of the cycle (Fig. 5). These observations are consistent with the results of Flamm andDonnelly (17), who showed that time-dependent Cl* and Ar* emission (at 250 kHz) have the samefunctionality in the plasma center but not in the plasma sheaths. In this case, it is concluded that Arcannot be used as an actinometer for determining Nc1 in the electrode sheaths.

Fig. 5. Ar* (left) and Cl* (right) emission lineshapes asa function of time during the (50 kHz) rf cycle1 mm above the powered electrode: at the voltagemaximum (momentary anode), top; at the voltageminimum (momentary cathode), middle; andtime-averaged (bottom) (Ref. 16).

Ar* EMISSION (X826.4nm) C EMISSION (X837.6nm)


C. Temperature Measurements

Spectrally resolved electron-impact emission can be used to determine rotational and vibrationaltemperatures providing the emission is short-lived and results primarily from direct excitation ofground-state species (1 8) . If chemiluminescence or electron-impact induced dissociation are sources foremission intensity or the excited-state collisionally relaxes on a time-scale which is fast compared to itsemission rate, the temperatures obtained will be valid for only the excited state. We will mention only afew recent studies relevant to plasma processing. Of course, it must always be remembered that theglow discharge is a non-equilibrium system and a single temperature is almost never adequate todescribe the energy distributions for all degrees of freedom of all atoms and molecules.

As a measure of carrier gas temperatures, optical emission from inert molecules such as N2 and COis probably reliable. Harshbarger and Porter (19) showed recently that N2 second positive emission(C3n —+ B3Hg) from a N2 discharge can be accounted for by direct electron-impact excitation ofx1- (v"=O): the vibrational intensity distribution is a convolution of the C3ll —X1 (v"=O)excitation and C3ll — B311g emission Franck-Condon factors. These same authors had used N2emission previously (20) to measure a rotational temperature of 595 ± 30K in an rf (13.56 MHz)discharge through N2. Oshima (21) determined rotational temperatures from CO emission, formed inC2F6/02 plasmas. He reports values for Tr which range from 400 to 600K depending upon gascomposition and applied power. DeBenedictis et a!. found similar results from CO emission excited inCO/He plasmas (22) . Although these authors report substantial vibrational excitation of CO, theyapparently deconvolved the observed vibrational band intensities with respect to only the emission andnot the excitation Franck-Condon factors.

Davis and Gottscho (23) verified the use of optical emission from N2 and N in determining Tr bycomparing optical emission measurements to laser-induced fluorescence (LIF) measurements(Section III). In N2 discharges, the two techniques gave consistent values for Tr, demonstrating thatemission line intensities can be used for ground-state rotational energy distribution measurements. InCCI4/N2 mixtures, Tr values from radical CC1 LIF and N2 were generally consistent except near cooledelectrodes where the CC1 temperatures were slightly colder, probably as a result of ion sputtering of coldadsorbates.

Optical emission may also be useful as a measure of the electron temperature. Again, the sameconstraints mentioned above apply here. D'Agostino et a!. (5) examined the intensity ratios of lines withdifferent energy thresholds to deduce that changes in the electron temperature were negligible as CF4was diluted with 02. In a similar fashion, Mehs and Niemczyk (24) used various Ar and Fe lines todeduce electron temperatures of 2OOO-5OOOK in a hollow cathode discharge.

D. Time-Resolved Emission

Recently, emission from rf discharges has been time-resolved in order to obtain dynamicalinformation on the excitation function (17,25,26). These studies are similar to those on ionization wavesin dc discharges, for which a more extensive review can be found elsewhere (27). The primarydifference is that while ionization waves in dc discharges stem from slowly evolving instabilities, in rfdischarges they are driven by high frequency applied fields. The experimental configuration used fortime resolution is like that for any optical emission experiment except that the photomultiplier tubeoutput is fed into a gated integrator which is triggered, in turn, by a reference waveform locked to theapplied voltage (Fig. 6).

In low frequency (250 kHz) Cl2/Ar plasmas, Flamm and Donnelly (17) found that the opticalemission from Cl*, Cl+ , Cl , and Ar* were in phase with the voltage waveform in the momentarypositive (anode) electrode sheath but lagged the voltage in the momentary negative (cathode) electrodesheath. The delay was attributed to the sheath transit time for ions, which eject secondary electronsupon electrode impact. The secondaries then excite emission as they traverse the sheath. Gottscho et a!.(26) made similar observations in N2 plasmas (at 20-90 kHz) except that no phase delay was apparentduring the negative cycle; this difference may be a result of the greater mobility of N (or N) ions inN2 compared to Cl (or Cli in Cl2 and Ar as well as the lower frequency range employed in the latterexperiment. Flamm and Donnelly (17) also noted that the emission was 100% modulated throughoutthe discharge because the electron energy distribution relaxes on a time-scale which is fast compared toone-quarter period of the applied potential (28).

At higher frequencies (>1 MHz), larger phase shifts and smaller modulation are observed (17,25).The phase shift was found to be dependent upon position from the electrode surfaces: the excitation wavepropagates away from the cathode and narrows to a pulse with FWHM of —10 nsec, much shorter thanthe applied voltage period (74 nsec for 13.5 MHz) (25). An explanation for the formation of this pulseis still wanting.


— ————TRIGGERw —THRESHOLD

fl241I I I I

zo

VARIABLEDELAY

LASERPULSE

GER

I LASER—INDUCED

—IziwII

—IzI—

FLUORESCENCE

1 I I I I I IILoIwICF-Iw<I-DIw(IJ

DELAY

TRIGGER

I I I I

GATE

3 4Fir)

0 1 2TIME (UNITS 0

In general, we have seen that optical emission is most useful for qualitative analysis and semi-quantitative characterization of energy distributions and excitation phenomena. In some instances,actinometry can be used to improve its quantitative utility.

III. LASER-INDUCED FLUORESCENCE

From the above discussion it is clear that emission spectroscopy has a useful role to play in plasmadiagnostics but that there are also very definite limitations upon the technique. Most particularly onedesires non-intrusive optical techniques which monitor directly the ground (or metastable) statepopulations of the active species. In principle laser induced fluorescence (LIF) is precisely such atechnique. Fig. 7 shows a schematic diagram of an experimental LIF apparatus. One passes a tunable

Fig. 7.

LIF schematic (Ref. 30).

PHOTO-MULTIPLER

Fig. 6.

Schematic timing diagram illustrating manner bywhich time-resolved optical emission and LIF arerecorded (Ref. 26).

THERMOCOUPLEPROBE

HEATER / COOL ERVARIABLEFREQUENCYGENERATOR

LASERBEAM

PUMP


dye laser through the plasma and collects laser induced fluorescence perpendicular to the laser axis.Typically one uses a pulsed dye laser and a gated detection system to reject the continuous backgroundemission from the discharge. An excitation spectrum of the species of interest is recorded by sweepingthe laser frequency and monitoring the intensity of laser induced fluorescence. Typically for plasmaexperiments the fluorescence is dispersed through a monochromator which serves as both a spatial andspectral filter to discriminate against the plasma background. When an excitation spectrum is recordeda relatively wide monochromator bandpass is employed; alternatively the laser frequency can be fixed onan excitation spectrum feature and the monochromator scanned with a narrow bandpass to record alaser-excited, wavelength-resolved emission spectrum. By employing such techniques the speciesmonitored can be unambiguously identified.

For such an experimental set-up the laser induced emission, If,of a reactive species can be related toits concentration En by an equation of the form (Eq. 2b), that we used to describe emission, i.e.

If=cf[r1 (6)

and in analogy with Eq. 3

af K ( Q (P,Ne) N1 () th (7)

oe () = cross-section for laser excitation of the ground statespecies r.

N€ (€) = number of laser photons available at energy .

The quantum yield function Q (P,Ne) is exactly the same as before.

At first glance it may appear that we have gained nothing by replacing Eqs. 2 and 3 respectively by 6and 7. However under most conditions Eq. (7) for a' greatly simplifies. The cross-section if () isdiscontinuous as a function of , thereby defining the line nature of the LIF excitation spectrum. Sincewe can externally control the number of laser photons and their frequency, i.e. , we can choose toevaluate af at the maximum of a particular line, i, in the spectrum. In that case

If = af Er] = K Q(P,Ne) Fri N Er]

K Fri N Er] (8)

where in the last equality Q(P,N) is assumed unity and Fri is the maximum oscillator strength for aparticular transition i of species r. In principle the quantity Fri is directly calculable (23,29,30) and K isdeterminable by calibration of the detector. Thus one has a direct measure of the concentration Er] ofthe ground state species r, essentially independent of discharge conditions, gas mixture, etc.

Eq. 8 is applicable for the volume of the reactor intersected by the laser and the field of view of thedetector. Such a volume can be made quite small, e.g. <1 mm3, thus one can probe different spatialregions of the reactor. In a similar manner the laser pulse is very short ( 10 nsecs), so the timeresolution of the technique is very high. The overall sensitivity of the LIF technique is also very high.Numerical evaluation of the quantities appearing in Eq. 8 leads to detectable signals for Er] as low as106/cm3 and less.

It should be pointed out that the concentration Er] given in Eq. 8 is the concentration of theparticular quantum state in which the transition originates. If the molecular internal states arethermally equilibrated and their temperature known, the concentration Er] may be related to the speciesconcentration in all levels using the well-known Boltzmann relation (23,30). By changing the dye laserfrequency, the relative populations of many quantum states can be measured. From this information,the extent of thermal equilibrium can be established and the temperature of the species determined.

It is usual for the translational and rotational degrees of freedom to be equilibrated for a givenspecies in a plasma. However, the degree of equilibration between ions and neutrals is still open toquestions. LIF is an excellent tool for measuring the internal rotational temperatures of the variousspecies directly in their ground states. As discussed above (Sec. II.C), emission spectroscopy providessimilar information for the excited states.

A. Concentration Measurements: Radicals

There are several reasons for wanting to know the concentrations of various radicals produced in rfplasmas. The most obvious reason is to be able to control the heterogeneous chemistry by changing theplasma composition and radical transport to device surfaces. Small radicals are suitable formeasurement by LIF in that Q is often nearly unity. To date, the most extensive LIF studies of radicals


in plasmas have been for CF2 and CC1 radicals in fluorocarbon (31,32) and chiorocarbon (12)discharges, respectively.

Hargis and Kushner (31) measured relative CF2 concentrations by LIF with a KrF excimer laser(248 nm). Since a fixed frequency laser was used, no knowledge of the internal energy distributions wasobtained but it is likely that the levels sampled were representative of the total CF2 concentration, whichwas examined as a function of applied power (at 13.56 MHz). C2F6 plasmas were found to generateabout four times as much CF2 as CF4 plasmas. In both systems, the CF2 density increased super-linearly with rf power.

In a separate LIF study, Pang and Brueck (32) measured CF2 concentrations in CF4/02/H2 plasmaswith a quadrupled Nd:YAG (266 nm) laser. Again, the internal energy distribution was notcharacterized. They found a similar dependence of [CF2I on rf power to that obtained by Hargis andKushner (31). In addition, they examined the dependence of [CF2] on pressure, frequency, 02concentration, and H2 concentration. As expected, [CF2] decreased as 02 was added to the feedstockbecause of oxidation reactions which form CO and CO2 and liberate free fluorine (33,34). Also asexpected, H2 addition caused [CF2] to increase because of abstraction reactions which result in HFformation (35). As the frequency was increased from 4 to 14 MHz, [CF2] was found to increasemarkedly most likely because of increased charge density at higher frequency (30).

Gottscho et a!. (12) used a tunable dye laser to monitor relative changes in CC1 radical density byLIF as a function of power, flow-rate, pressure, electrode temperature, feedstock composition, andposition between the electrodes. Because of the dye laser's tunability, it was possible to select rotationaltransitions in the A2—X2H band system which are temperature insensitive and therefore trulyrepresentative of the total CC1 concentration. The spatial variations in concentration as a function ofpressure show clearly that CC1* emission is a poor measure of [CC1I (Fig. 8). At low pressure (155 mTorr = 20.6 Pa) (Figs. 8a and 8b), where the mean lifetime for ground-state CC1 is comparable to orlonger than the diffusion time, the [CC1] spatial profile is dictated primarily by the thermal gradientimposed across the gap (12): the left electrode was heated to 300CC while the right electrode waswater-cooled. By contrast, CC1* emission exhibits maxima in both electrode sheaths and is relativelyinsensitive to thermal gradients. At higher pressure (106 Pa, Figs. 8c and 8d), both CC1 concentrationand emission profiles show strong maxima in the electrode sheaths and only weak signal strengths in theplasma center. These changes as a function of pressure are readily understood when one considers thedifferences in lifetime for the excited A2 and ground X2H states of CC1. The A2z state has a radiativelifetime of 105 ± 3 nsec and a quantum yield which changes from 90% to 60% when the pressurechanges from 20.6 to 106 Pa (36). Because the X2H density is low (<10'°moleccm3) and the A2

(a) l55mTorr 30 (b) l5SmTorr

__________________ C /TTTTTTIIITI,(c) (dl

800mTorr (1)30 800mTorr

:M5IONAXIAL POSITION (mm) t AXIAL POSITION (mm)

COLD HOT COLD HOT

Fig. 8. CC1 emission (left) and LIF (right) vs. positionfor two different pressures in a CC14 discharge:top, 20.6 Pa (0.155 Torr); bottom, 106 Pa (0.800Torr). The electrode on the left is heated to300 C while the right electrode is water cooled.


lifetime is short, CC1* emission arises mainly from electron-impact dissociation of CC1 species and ismostly a measure of the overall rate of CC1 formation. The X2H state, on the other hand, has a lifetimeof 5O sec at 1 Torr (12,14), which is most likely governed by reactive collisions such asCC1+CC14 — CC12+CC13. At higher pressure, the X211 state cannot diffuse far from its site offormation without being lost by reaction. Thus, the emission and concentration profiles are similar athigher pressure since both reflect the CC1 formation rate.

The observation that CC! radicals are formed in the electrode sheaths is consistent with a model forCC14 decomposition by dissociative electron attachment (1 2) . The CCI4 electron attachment crosssection is strongly peaked only at energies below 10 eV, with the largest peak at zO eV. Slow electronsare readily captured by CC14 (and presumably by CClX species as well) to form metastable negative ions(e.g. CC141 which subsequently dissociate and produce radicals such as CC13 or CC1 and stable negativeions such as Cl (37-40). The slowest electrons in the plasma can be found close to the electrodesbecause this is where the potential is most negative: secondary electrons formed by either ionizingcollisions or ion-impact of the electrodes have not yet undergone acceleration by the field; similarly,plasma electrons have been slowed by the large negative sheath potential. The end result is that radicalssuch as CC1 are formed predominantly in the sheath but will be confined there only when the reactiveloss rate exceeds the diffusion rate.

In another application of LIF to plasma diagnosis, Uchino et a!. (41) monitored the n=2 and n=3levels of hydrogen by exciting the Balmer alpha transition and detecting plasma-induced emission,respectively. They used these measurements to deduce the amplitude of ion-acoustic turbulence across apotential step in a low pressure dc discharge through H2. This experiment is noteworthy for two reasons:(i) it illustrates that LIF can be used to measure highly excited but metastable level populations ofatomic radicals and (ii) that LIF can be detected sensitively in the presence of not only intense plasma-induced emission but also intense laser radiation of the same wavelength.

B. Concentration Measurements: Ions

Donnelly et a!. (30) looked at the time-averaged concentration of Cl ions in a Cl2 plasma as afunction of frequency, power, pressure, feedstock composition, and distance from the electrodes. Thedependence of [Cl] on frequency was of primary interest, since frequency can play an important role inetching and deposition (30,42,43). As frequency was varied from 13 MHz to 0.01 MHz, a minimum inthe Cl density, most pronounced in the sheath at low power density, was observed at 1 MHz(Fig. 9). This frequency roughly corresponds to the ion plasma frequency, above which the ions can nolonger respond to the applied rf potential. At low frequencies, (<1 MHz) ions can follow thefield (26,43) and impact electrode surfaces with sufficient energy to generate secondary electrons, whichare then accelerated and produce ionization as they cross the sheath. As the ion plasma frequency isapproached, however, the sheath field decreases, ion-impact no longer produces many secondaryelectrons, and the charge density decreases. Above 1 MHz, the charge density begins to increase again,but the mechanism by which this occurs is not well understood.

(I)I.-z

cc

+c'j

Fig. 9. C1 (LIF) vs. freq. (Ref. 30).

FREQUENCY(MHz)

100


In order to monitor the dynamics of low frequency plasmas, Gottscho et a!. (26) measured N andCl ion densities by time-resolved LIF. In these experiments, the dye laser used to probe ground statepopulations is pulsed synchronously with the applied rf potential in order to sample the instantaneous ionconcentrations anywhere in the gap (Fig. 6). In the plasma center, the N and Cl populations werefound to be only weakly modulated when pure N2 and Cl2 'plasmas were examined. The extent ofmodulation was used to estimate 1st order ground-state lifetimes as a function of pressure for Nand

Cl: at 0.3 Torr, 30-35 psec for both ions. These times probably represent loss by ambipolar diffusion.When 10% Cl2 was added to the N2 plasma, the N lifetime decreased to 1.5 tsec while the Cllifetime increased. This indicates that the charge exchange reaction N+ Cl2 — Cl+N2 occurs readilyin N2/Cl2 plasma mixtures (k - 6 x 106 sec1 Torr").

In the electrode sheaths, the ionic response to the applied potential was characterized in terms of anoscillating sheath front. At any position in the sheath, the ion population first grows because ofelectron-impact ionization of neutral N2 or Cl2; but, as the sheath front propagates out from theelectrode surface, ions are rapidly accelerated toward the surface while electrons are repelled toward theplasma center (Fig. 10). Over the frequency range studied (20-90 kHz), the sheath propagation velocity(determined from the line in Fig. 10) was found to increase linearly with frequency so that the sheathfront reached the distance of maximum sheath expansion, d, in one-half period; over the same frequencydomain, d was observed to be a function of pressure and gas composition only. For N2 plasmas andlower pressures, d was larger than for Cl2 plasmas and higher pressure. The pressure dependence wasalso observed by Donnelly et a!. (30) in their time-averaged LIF experiments and is consistent withreduced charge mobility and increased charge density at higher pressures. Cl2 plasmas exhibit smallersheaths for similar reasons: the higher electronegativity for Cl2 reduces negative charge mobility and thelower ionization potential for Cl2 leads to higher charge densities.

Fig. 10.

N LIF conc. vs. time during the rf cycle forfour positions in the sheath. For comparison theapplied voltage waveform is shown at the top andplasma-induced emission is shown for a position0.75 mm from the electrode at the bottom.

C. Energy Distribution Measurements

Because of the tunability and resolution of LIF, it is nearly ideal for measuring internal energydistributions of ions and radicals in rf plasmas (23,30,44). Most measurements have been made on therotational energy distributions and in every case reported so far, the distributions could be representedby a single rotational temperature, Tr. Donnelly eta!. (30) saw little dependence of the Cl rotationaltemperature on distance from the water-cooled electrodes or on applied frequency (0.25 or 13 MHz);

Eoci

0E>.

(I)zU0

.fNcJx

LiiI-F-U)

0z0

z

0 0.5 1.0 1.5TIME (UNITS OF 'iT)

20


they measured an average Tr of 350K. Davis and Gottscho (23) investigated the influence of surfacetemperature and rf power on the CC1 rotational temperature profile across the electrode gap in a Cd4plasma at low frequency (55 kHz) . They found little dependence of Tr 01 position except near theelectrodes, where gradients as large as 102K mm1 were observed. The dimensions of the thermalgradient are consistent with known mean free paths for rotational relaxation; the magnitude of thegradient and the temperature in the plasma center are governed by both electrode temperature and rfpower density.

No detailed LIF studies of vibrational energy distributions in plasmas have yet been reported,although hot bands have been excited in CC1 (23) and N (26). The results obtained suggest that thevibrational "temperature' is substantially greater than the rotational temperature as is expected fromtheir different rates of relaxation.

Nor have detailed LIF studies of translational energy distributions been reported yet. There are tworeasons for this: (i) for radicals (or ions in the discharge center), a narrow frequency, cw laser would berequired to accurately measure Doppler line shapes at temperatures less than 1000K but all LIF studiesof plasmas to date have utilized pulsed lasers with linewidths comparable to or greater than the Dopplerwidth; and (ii) for ions, the broader energy distributions expected to occur in the sheath (45), have notbeen observed by LIF because the steady-state ion concentrations are so low (<106 molec cm3) as aresult of acceleration by the sheath field (26,44). While the first limitation can be circumvented, thesecond limitation is more fundamental. Measurements of the ion translational distribution in the sheathrequire techniques other than LIF, such as laser optogalvanic spectroscopy (LOG) (44). It should bementioned that a high resolution cw laser has been used to measure the absorption line strength andshape of OH in an H2O discharge, which give both Tr and T (46).

IV. OTHER TECHNIQUES

A. Opto-Galvanic Techniques

A diagnostic technique very closely related to LIF is laser optogalvanic spectroscopy (LOG). Theeffect was first discovered in discharges over 50 years ago, with an atomic lamp standing in for the laser.However, the development of tunable lasers has greatly widened (47) the applicability of the technique.

A schematic diagram of a typical LOG apparatus is shown in Fig. 11. Just as in LIF, an excitationspectrum of a species of interest is obtained by scanning a tunable dye laser through one of the species'electronic absorptions. However rather than detecting fluorescence created by the absorption of laserphotons, optogalvanic signals result from changes in the impedance of the discharge in response to theabsorption of laser radiation. Usually the impedence change is monitored as a change in voltage acrossa DC discharge as shown in Fig. 11. For rf discharges, laser induced impedence changes are detectedeither by monitoring the response of the driving circuit or by a separate RF antenna.

GAS GASOUTLET INLET

LJ \LASER

Fig. 11. LOG apparatus.The spectra of numerous atomic species, ions, free radicals, and relatively unreactive feed gases, have

been obtained in discharges by the LOG technique. The chief disadvantage of the LOG experiment as aplasma diagnostic is in the relation of the observed LOG signal strength to the concentration of thecarrier species. If one writes an equation of the form of 2 or 6 then the functional form of theproportionality factor a is not even clear. In simplest terms the absorption of a laser photon with theconcommitant change of electronic state in the absorbing species causes the observed impedence changein the discharge. However the detailed relationship between photon absorption and impedance change isgenerally very complex.


A typical mechanism would be the production of an excited state of a neutral atom or free radical;this excited state could be more easily ionized leading to more ions. more discharge current, and lessapparent discharge impedance. Conversely the laser excitation could lead to a less easily ionized state,giving fewer ions and thus a higher impedence. Indeed signals of opposite polarity are often observed inLOG spectra and a given transition can change polarity as a function of discharge current or pressure.Usually there are many competing processes in the plasma and the absorption of laser photons simplyshifts the balance.

As the above discussion implies LOG is likely an unreliable tool for quantitative determination ofspecies concentrations. However it may still be quite useful as a plasma diagnostic because the veryprocesses — ionization, electron attachment, photodetachment, and recombination, which are responsiblefor the generation of LOG signals — are the same processes that sustain and control the discharge. Bystudying the LOG signals in concert with other information, say LIF signals from the same species, itmay be possible to unravel the complex kinetics governing the plasma.

Another instance of the usefulness of LOG spectroscopy is the situation where other spectralsignatures of a species are absent but LOG signals are present. An important example of such asituation is the cathode sheath region of an rf discharge. As pointed out earlier LIF detection of N iseasily seen in the center of such a discharge but is not detectable in the cathode sheath because thehigher ion velocities there limit its concentration. Walkup, Dreyfus, and Avouris (44) have recentlyreported LOG signals for N in the sheath region of the discharge, while the LOG signal drops belowdetectability in the center of the discharge. It is argued that the high sensitivity for LOG detection ofN in the sheath results from an effective increase in ion mobility due to a smaller cross section forcharge exchange collisions in the excited state.

The obtainment of the LOG signal of N in the sheath allows a measurement of its rotationaltemperature which is warmer than in the rest of the plasma. Also Doppler broadening was detected forN moving along the electric field lines (44).

B. tnfrared Techniques

Surprisingly few applications of infrared spectroscopy to plasma processing have been reported.Fourier transform infrared (FTIR) emission spectroscopy has been used in characterizing low pressuresilane (48) and air (49) plasmas but JR emission is subject to the same limitations as optical emissionand is useful mostly for qualitative analysis. Knights et al. (48), using a globar source and FTIR,measured the spectra of SiH4 and SiH in rf (27.3 MHz) plasmas. They made no attempt to obtainspatial information and were unable to detect polyatomic radicals. such as SiH2 and SiH3. Fromrovibrational line intensities, they determined the following vibrational and rotational temperatures: forSiH4, T=85OK and Tr=300K; for SiH, T=2OOOK and Tr=484K. The non-equilibrium betweenvibration and rotation is consistent with LIF experiments and is not surprising since vibrationalrelaxation generally proceeds more slowly than rotational relaxation. Nor is it surprising that the SiH4parent is cooler than the SiH radical, which is formed predominantly by electron-impact dissociation. Inanother JR absorption experiment the decomposition of CF4 and C2F4 was investigated (50).

Based on recent experiments, we expect that infrared laser techniques will make a significantcontribution to plasma diagnosis. Hirota et a!. (51-53) have used tunable diode laser spectroscopy(TDLS) to record the absorption of several free radicals (e.g. CC1, NS, and CS) created in ac plasmas.The concentrations of these species were not measured, but they are likely to be as low as I 0-lO10moleccm3 (12). Very recently, DeJoseph et a!. (54) used TDLS to measure SiH4 concentrations(—..10'2-lO14moleccm3) downstream from a dc discharge (i.e. ex situ). Fluorine (55) andchlorine (56) atoms have also been detected by TDLS but, again, only in ex situ experiments. Stantonand Koib (55) demonstrated a detection sensitivity for the electric-dipole forbidden transition

—2P112 in F of 1015 atoms cm3 and suggested that this could be lowered to —1012atoms cm3

by using multi-pass optics and derivative detection.

Oka et a!. (57) used TDLS in situ to measure Doppler shifts for different rotational and vibrationallevels of ArH ions created in a dc glow discharge (Fig. 12). From these shifts, drift velocities weredetermined, which in conjunction with probe measurements of the axial electric field, provided anestimate of the ionic mobility. A similar type of measurement was made by Saykally et a!. (58,59)using a color center laser to detect HCO+ and HNN+ ions, whose motion was modulated by a lowfrequency (1-8 kHz) driving potential. In these experiments, the primary emphasis was on the rotation-vibration line positions; the ion drift velocity was used to enhance the detection sensitivity with respect tounaccelerated neutrals.

In another JR laser experiment, Farrow and Richton (60) used a CO2 laser to monitor the rotationaltemperature (675 ±75K) and degree of dissociation (5-20%) of BC13 in an rf (13 MHz) discharge.

PAAC 56:2—C


DRIFT VELOCITY GAS( INLET

4,

All of these experiments clearly demonstrate the capability of IR techniques for in situ monitoring ofthe concentrations and energy distributions of radicals, ions, and parent gases found in low pressureplasmas. Although spatial information has not been obtained yet, there is no reason why tomographictechniques could not be employed to circumvent the line-of-sight limitations of absorptiQnspectroscopy (61 ,62) . Application of TDLS to plasma processing seems particularly promising becauseof the compactness and relative economy of diode laser sources.

C. Raman Techniques

Raman spectroscopy has yet to be exploited as a plasma diagnostic tool despite its inherently highspatial resolution and ubiquitous use elsewhere. Like LIF, both spontaneous and coherent Ramantechniques entail detection of scattered photons; but, unlike LIF, Raman scattering does not rely uponcreation of a fluorescent excited state and is not sensitive, therefore, to excited state quenching.

Several recent applications of Raman scattering to detection of atomic radicals are encouraging (63-67) Cummings and Aeschliman (63) measured the spontaneous Raman scattering cross section for the2p3/2 —p 2112 fine-structure transition in fluorine. They used a cw Ar+ laser at 488 nm (1.1 W),focussed into a heated monel cell containing an equilibrium mixture of F and F2. The rotational Ramansignal from F2 was used to determine the temperature, which in turn was used to determine the F atomdensity (.1015l017 atoms cm3 for T=600 to 850 K). In an ex situ experiment, Moore (64) utilizedcoherent anti-Stokes Raman scattering (CARS) (see Ref. 65) to detect the corresponding fine-structuretransition in Cl, formed upstream in a microwave discharge. In this experiment a Nd:YAG laser wasused as a pump for both the CARS signal and for a dye laser whose output is used as the tunable Stokesfrequency (65). Moore also detected nonresonant CARS from Cl2 but saw no change when thedischarge was switched on or off. The Cl atom density was measured by titration with NO or Br2 usingthe disappearance of the CARS signal as an end point indicator. A detection limit of (4.5 ±1.7) X 1015

atoms cm3 was reported but this might be improved by two orders of magnitude by using narrowerbandwidth lasers (64). Very recently, CARS has been used to measure the degree of SiH4decomposition in a DC glow discharge at 0.5 Torr as a function of position between the electrodes andas a function of pressure (67).

Hargis (68) has shown that the sensitivity of spontaneous Raman scattering can be made comparableto that for coherent Raman techniques by simply using an UV source. He demonstrated this by using aKrF excimer laser (248 nm) to induce Raman scattering from N2 at partial pressures as low as i0Torr.

Certainly, the primary reason why Raman spectroscopy has yet to be used extensively as a lowpressure plasma diagnostic is its lack of sensitivity. However, we anticipate that Raman techniques willprove useful in measuring the degree of dissociation and temperature (69) of both feedstock constituentsand abundant (>1014 atoms cm3) radical atoms.

D. Multiphoton Techniques

It is appropriate to return to the species, with which we began our discussion, the light atoms, F and0. Fig. 13 shows a generic energy level diagram appropriate to all the reactive light atoms in the firsttwo or three rows of the periodic table. The key feature of this diagram is the very large gap in energybetween the ground and first excited state, followed by a large number of levels between the first excitedstate and the ionization limit. It is between these latter states that the transitions occur which were usedfor the emission diagnostic of the CF4-O2 plasma.

In our discussion of that work we pointed out both the importance of the emission diagnostic for

Fig. 12. Diode laser apparatus (see Ref. 57).


understanding the CF4-02 plasma and the limitation in the technique, which centered around theuncertainities in ae the proportionality "constant" between concentrations and emission intensities and themode of production of the excited state. Because of these uncertainities and the importance of lightatoms for plasma processes, it may seem strange that a more "quantitative" diagnostic, like LIF, has notbeen applied to light atom detection. Fig. 1 3 gives the obvious answer, all transitions in such atoms

IONIZATION CONTINUUM

///2///////;//Z///

EXCITEDSTATEMANIFOLD,

AE 4O,OOOcm SPONTANEOUS VISIBLEOR NEAR-IR EMISSIONhi' >10,000cm1,xcI000nM

LOWEST ALLOWED ONE LOWEST ALLOWEDPHOTON TRANSITION 2-PHOTON TRANSITION

hzy8O,000cmH hl/2QhZ'2b 50,000cmtX1<I25nM hl/20,hlISb� 200aM

Fig. 13. Schematic energy level diagram for "light" atoms.On the left hand side a one-photon transition isillustrated, on the right a two-photon transition.This energy level diagram applies to H, N, 0, C,F, Cl, Br and other "light" atoms. The onlyexception to the stated energy intervals is for Fatoms (Ref. 74).

from the ground state lie in the extreme UV spectral region, well beyond the reach of today's tunablelasers. The key to the solution of this problem, as shown on the far right of Fig. 13, is the use of twophotons, rather than one, to reach one of the lower excited states. The requirement on the laserwavelength is then typically shifted into the 200-300 nm range, a difficult but not impossible region forthe generation of intense, tunable laser outputs.

This 2-photon LIF has presently been successfully applied to the detection of the light atoms, H(70), N (71), 0 (72,73), 5 (74) and Cl (74). These experiments bring all the traditional advantages ofLIF probes, high temporal and spatial resolution coupled with reliable relations between signals andconcentrations, to the detection of the light atoms. So far the successful experiments have been ex situin nature, i.e. the 2-photon LIF of the atoms was accomplished downstream of the discharge source.Experiments are presently underway aiming towards in situ 2-photon detection and monitoring of thesesame atoms.

V. CONCLUSION

It should be clear by now that no single diagnostic technique is adequate to characterize plasmachemistry or to provide process control. Optical emission is primarily useful for characterizing theelectron energy distribution and excitation function. LIF and multi-photon techniques are most usefulfor sensitive, spatially resolved detection of atoms, radicals, and ions; opto-galvanic techniques cancomplement the LIF techniques in regions of the plasma, such as the sheaths, where LIF isunobservable; IR and Raman techniques are well suited for detection of polyatomic, non-fluorescentmolecules such as the freons, which comprise the feedstock in many plasma processes.

A simple example may serve best to illustrate the as yet unfulfilled promise of optical diagnostics inthe plasma processing environment. When organic materials, such as polyimide or photoresists, areetched in an 02 plasma, anisotropy is only observed at low pressure, regardless of applied power (75,76).These observations appear to be in conflict with current theories that line profiles in ion-assisted etchingare controlled primarily by the ratio of the sheath electric field to the total gas density, ES/N (45).Since the rf power should be approximately proportional to IEsI2 (77), it is curious that anisotropicorganic etching cannot be achieved at high power density and high pressure. Basically two explanations

h1120

GROUND STATE


can be offered for the discrepancy between theory and experiment. First, if an increase in rf powermerely expands the plasma volume without increasing the power density, the measured rf power is nolonger a measure of the sheath field. Secondly, if the sheath field and ion flux to the organic surfaceincrease with rf power but the radical 0 atom flux increases by much more, then the ion-assistedchemistry, which is responsible for anisotropic etching, will become relatively less important than theneutral chemistry, which is responsible for iostropic etching. If higher anisotropic etching rates aredesired, then greater control of either the ion vs. neutral chemistry or the plasma volume or both athigher pressure is required. But how can this control be achieved if we don't know which explanation iscorrect?

Clearly, optical techniques can come to the rescue. For example, spatially-resolved optical emissionor LIF would be a good means to evaluate the plasma volume. The sheath electric field couldconceivably be measured by a TDLS Stark shift or an LIF Stark intensity measurement. Theconcentration of 0 and 0+ could be measured by two-photon LIF or by emission actinometry. The Oconcentration could be measured by one-photon LIF while the 02 concentration could be measured byRaman techniques. Once such experiments were performed, the explanation for the discrepancy betweentheory and experiment should be obvious. Given an explanation, let us say the neutral flux increases bymuch more than the ion flux, we could try to achieve better process control by adding a neutralscavenger and thereby suppress the neutral-surface reaction and increase the anisotropy. The effects ofadding the scavenger could be monitored in real time by the optical diagnostics described above.

Measured concentrations, fluxes, and fields in rf plasmas can be compared to results of kinetic modelcalculations (78,79) to obtain a better understanding of glow discharge processes. In the example citedabove, it is possible that the theory would be found wanting in which case we would have improved notonly our process control but also our fundamental understanding of plasma-surface chemistry.

REFERENCES

[ii W. R. Harshbarger, R. A. Porter, T. A. Miller, and P. Norton, Appl. Spectrosc. 31, 201 (1977).

[2] C. J. Mogab, A. C. Adams, and D. L. Flamm, J. Appl. Phys. 49, 3796 (1978).

[3] D. L. Flamm, Solid State Technology 22, 109 (April 1979).

[4] J. W. Coburn and M. Chen, J. App!. Phys. 51, 3134 (1980).

[5] R. d'Agostino, F. Cramarossa, S. DeBenedictis, and G. Ferraro, J. Appl. Phys. 52, 1259 (1981).

[6] R. d'Agostino, V. Colaprico, and F. Cramarossa, Plasma Chem. and Plasma Proc. 1, 365 (1981).

[7] R. d'Agostino, F. Cramarossa, V. Colaprico, and R. d'Ettole, J. AppI. Phys. 54, 1284 (1983).

[8] R. d'Agostino, F. Cramarossa, and S. DeBenedictis, Plasma Chem. and Plasma Proc. 2, 213(1982).

[9] V. M. Donnelly, W. C. Dautremont-Smith, D. L. Flamm, and D. J. Werder, J. Appl. Phys. (inpress).

[10] D. E. Ibbotson, D. L. Flamm, and V. M. Donnelly, J. App!. Phys. (in press).

[ii] H.-J. Tiller, D. Berg, and R. Mohr, Plasma Chem. and Plasma Proc. 1, 247 (1981).

[12] R. A. Gottscho, G. P. Davis, and R. H. Burton, Plasma Chem. and Plasma Proc. 3, 193 (1983); J.Vac. Sci. Technol. Al, 622 (1983).

[13] J. P. Simons and A. J. Yarwood, Nature 187, 316 (1960).

[14] R. A. Gottscho, unpublished results.

[is] T. Ogawa and M. Higo, Chem. Phys. Lett. 65, 610 (1979), and references therein.

[16] R. A. Gottscho and V. M. Donnelly, to be submitted to J. Appl. Phys.

[17] D. L. Flamm and V. M. Donnelly, Bull. Am. Phys. Soc. 27, 97 (1982).

[18] G. Herzberg, Molecular Spectra and Molecular Structure I. Spectra of Diatomic Molecules, (vanNostrand Reinhold Co., New York, 1950).

[19] W. R. Harshbarger and R. A. Porter, App!. Spectrosc. 35, 130 (1981).

[20] R. A. Porter and W. R. Harshbarger, J. Electrochem. Soc. 126, 460 (1979).

[21] M. Oshima, Jpn. J. App!. Phys. 17, 1157 (1978).

(6L61) 01717 'L9 °] S/q 'IIèI )I U jDU1 SO!AIU 1 d [9] (0861) L99 'L Si'4d f 'qioN 1 D PU1 UouS 3 •v cc}

(Z861 'su1301J0 MM) 86I 001d 'puOd •�1 •U PU 'UOpp1OSi19 •v ''f 'qdOSOf(J V 3 ft] (1861) 9Tt' 'c osoJDodS I°IAI •f '12OJiH 1 UE 'IEs.J ) 'PW1A •D

. (086 ) 89 'fry •Ds0J2DOdS jOJi\J •f '1°!H E1 PU 'PWA •D '!M •)I 'iqon1M) •N 'J1LUflSJ,,\ •)I [cc]

(6L61) £O 'frL osoJoodS jOJAj f 'ETOJIH •1 pU13 PEWA •J [Is]

(86I) cj7j7 '9E DodS 1ddy 'aiiiiojS H pu-e 'OZUIH a 'l1d Ti H [Os]

(61) g ' dj jddy 'UOJ2S f P-' '11O21Od JA,j 'UOSSUUjOf j 'uijdsj IAI 'UOSU1H d '!IUS •H [6i7]

(z861) It '9L SJ1d woqj f 'ijiAqDUon9 9 pui 'Uwod f '1uS JA •d f 'SIU) 3 f [817]

(E861 qJ) 117 'SflOOJ .10SWj 'JOU2OJ 1 3 UU osoqo )J 3 [Li7]

(LL6I) 66 '6F 3fl AOJ SA4d 'J8U!11!N )I Q pU U13M 3 3 [917]

(E861) 17j71 I 'O1 '°°S 1OP0.I2DE1 f 'UiMOJU 1 3 [17]

(E861 OUflf) sso.id U! 'TO AOJ Scq pui 'yçjj iDi.isqv 'pj,'j 'oJOLUu1 'E861 'OZ-L1 JcUJA 'SO!1dO-OJDOjJ pU SJOSUj UO JU03 'SUflOAV d UU 'SflJJOJU M J 'dn)IIUM 1 1 [1717]

(1861) 1790L 'ç sq iddV f '°°"1 H 1 [E17]

d '(0861 'S!flOj

iS) •5 .I1I 'qU8oAJ f 3 pu .iOSOUJ 9 )J po 'DOId 1WSjd dw1c5 Doid 'niq H [17]

(861) 906 '[ SIc1J jddy ç Udf '!)pZ)jy JAJ pUI '1)jOUIfljAJ N 'UpOUO5 A 'UO)j N 'UpoUJi\J J,\J 'OU!qof )I [117]

(0861) £ '9 dSIod 12P°H UOUAUJ 'noioqdosuqj 9 ] [017]

(0861) osc 1u0q3 sqj SO8UOSUn 'L '1o2 unii Fl 'Ja8JoqUojjj 'UU1UOUflO4D5 ri Fl [6E]

U!O1041 SODUOJOJOJ

UU '(cL6l) l7cE ' SIcqd 111043 'JojpU!q05 pu 'sopqnopo2suq3 v v 'Sojflqo5 1 [SE]

(6961) I6LI 'ic sqd uq3 !' 'q!oN UU 'o8ioo9 }j 'qJOMUoM 1 A\ [LE]

(861) 86zc Sq 111043 f 'S!A(J d 0 UU 'UOIfll H j 'OqDsflo9 'V èI [9E]

(6L61) t6c9 'oc SJcqd iddV f 'JSU!1OmS 9 pUU 1 PS00.IJ. 'V El [SE]

(1861) LE 'I °°1d LUS1Id pUU LUO4J UWS1jd 'Lu1uULl '1 a [17E]

(6L6I) 8617 'oc SA4d jdd'V f '1ULUUjJ 'i a pui lc)jSU!j0w5 9 [EE]

191 d '(E861 ')1IOA MOM) 'pU1jjOH 4JOM 'J0qssojqo5 }1 H UU f 1 S 'jOOSo IAJ

J p0 'a9htdJ JOJdflpUOd!WdS iof U!SdOJJ /VdltWfl/dOJOl/J pUn S3iJSOUV!cJ J7S'V7 U! 'U!q01 UUJSI4d H/ZO/J3 JO SOi2SOUiQ 0OU0OS0JOfljJ p0OflpUI-J0SW f s pUU UUd S [E]

(86T) 6LL 'Ofr 12°1 SJcqd jddv 'IOU4Sfl)I f j,% pUU If 'S!8IUH f d [If]

(861) LIS '[ jOUqOOj !05 °A f 'SU!jjO3 9 pU1 'U1.UUjJ ' ( 'Jc0UU0(J F\1 A [Of]

(LL6I) 617f ' 111043 S/4d Aoj UU'V 'í0SW)j '1 f [6]

(EL61) LLZ 'I mpUnJ LUO4J 8u pUl 'noppicj j j pui 'pU1pj!9 I 1 '111111U1J ] Q [8fl

LoI-c9 dd '(8L61 'JOA MOM) SSOJd 0!LU0p1D'V 'w1S H UU 4SJIH M

y.j ícq po ' /04 sdU0Jpd/q s'nosn9 U! 'soSIUqDs!Q M0j9 U! S0MM UO!P!Z!UOj,, 'u0pp!OSJU9 'V [Lfl

SA4d jddy f 02 ponuiiqns oq 01 's!AUU d 9 U1 'lcjlOUUOQ JAI A '1 (3 'U01Jfl H )J '040s1109 'V d [9]

(f86I)

uz 'c sIqd idd'V f 'SU.ION 3 J pUl3 'pUUAO(3 9 'UOSj0'V J f 'Jf 'InqsoJI\J ) J '1cUSO) 9 [c (1861) 99 'c osoJloOd5 iddv ')jIcZD1U0!M JAJ j pUI3 S4OJA,J JAJ a [17fl

(f861) 080E 'frc SJcqd jddy f 'O4DS1TOD 'V >J pui S!M!(3 d 9 [EZI

(861) L17Z 'IL Sc4d 111043 'USSOI1UJ1!I3 J U1! 'OU!1SO'Vp J 's!o!poUoqoa S [ZZ]

LO sTsouip uisid ut snbiuqzui iuziEdO


[57] N. N. Haese, F. Pan, and T. Oka, Phys. Rev. Lett. 50, 1575 (1983).

[58] C. S. Gudeman, M. H. Begemann, J. Pfaff, and R. J. Saykally, Phys. Rev. Lett. 50, 727 (1983).

[59] C. S. Gudeman, M. H. Begemann, J. Pfaff, and R. J. Saykally, J. Chem. Phys. 78, 5837 (1983).

[60] L. A. Farrow and R. E. Richton, unpublished.

[61] B. S. Choi and H. Kim, App!. Spectrosc. 36, 71(1982), and references therein.

[62] M. Deutsch and I. Beniaminy, J. App!. Phys. 54, 137 (1983), and references therein.

[63] J. C. Cummings and D. P. Aeschliman, Opt. Comm. 31, 165 (1979).

[64] D. S. Moore, Chem. Phys. Lett. 89, 131 (1982).

[65] A. C. Eckbreth and P. Schreiber, Chemical Applications of Nonlinear Raman Spectroscopy, ed.A. B. Harvey, Academic Press (New York, 1981), Chapter 2.

[66] Spontaneous Raman scattering has also been used to detect oxygen atoms in high pressure flamesby J. Dasch and J. H. Bechte!, Opt. Lett. 6, 36 (1981).

[67] N. Hata, A. Matsuda, K. Tanaka, K. Kajiyama, N. Moro, and K. Sajiki, Jpn. J. App!. Phys. 22,LI (1983).

[681 P. J. Hargis, Jr., App!. Opt. 20, 149 (1981).

[691 A. V. Voskanyan, S. B. Leonov, V. E. Mitsuk, and Y. A. Rusanov, Soy. Tech. Phys. Lett. 7, 481(1981).

[70] J. Bokor, R. R. Freeman, J. C. White, R. H. Storz, Phys. Rev A 24, 612 (1981).

[71] W. K. Bische!, B. E. Perry, and D. R. Cros!ey, Chem. Phys. Lett. 82, 85 (1981) and App!. Opt.21, 1419 (1982).

[72] M. Alden, H. Edner, P. Grafstrom, and S. Svanberg, Opt. Comm. 42, 244 (1982).

[73] P. Brewer, N. Van Veen, and R. Bersohn, Chem. Phys. Lett. 91, 126 (1982).

[74] M. Heaven, T. A. Mi!!er, R. R. Freeman, J. White, and J. Bokor, Chem. Phys. Lett. 86, 458(1982).

[75] F. Y. Robb, E!ectrochem. Soc. Meeting, San Francisco, Ca., May 8-13, 1983, abstract# 176.

[76] B. R. Soller, R. F. Shuman, and R. R. Ross, E!ectrochem. Soc. Meeting, San Francisco, Ca.,Abstract# 180.

[77] C. B. Zarowin and R. S. Howath, J. E!ectrochem. Soc. 129, 2541 (1982).

[781 M. J. Kushner, J. App!. Phys. 53, 2923 (1982).

[79] D. Ede!son and D. L. F!amm, 7th !nternationa! Symposium on Gas Kinetics, Aug. 23-27, 1982,Göttingen, Germany, Abs.# D48.

techniques in plasma diagnostics richard a. gottscho terry...

Documents